CLUES: Few-Shot Learning Evaluation in Natural Language Understanding

This repo contains the data for the NeurIPS 2021 benchmark Constrained Language Understanding Evaluation Standard (CLUES).

Leaderboard

We maintain a Leaderboard allowing researchers to submit their results as entries.

Submission Instructions

• Each submission must be submitted as a pull request modifying the markdown file underlying the leaderboard.
• The submission must attach an accompanying public paper and public source code for reproducing their results on our dataset.
• A submission can be toward any subset of tasks in our benchmark, or toward the aggregate leaderboard.
• For any task targeted by the submission, we require evaluation on (1) 10, 20, and 30 shots, and (2) all 5 splits of the corresponding dataset and a report of their mean and standard deviation.
• Each leaderboard will be sorted by the 30-shot mean S1 score (where S1 score is a variant of F1 score defined in our paper).
• The submission should not use data from the 4 other splits during few-shot finetuning of any 1 split, either as extra training set or as validation set for hyperparameter tuning.
• However, we allow external data, labeled or unlabeled, to be used for such purposes. Each submission using external data must mark the corresponding columns "external labeled" and/or "external unlabeled". Note, in this context, "external data" refers to data used after pretraining (e.g., for task-specific tuning); in particular, methods using existing pretrained models only, without extra data, should not mark either column. For obvious reasons, models cannot be trained on the original labeled datasets from where we sampled the few-shot CLUES data.
• In the table entry, the submission should include a method name and a citation, hyperlinking to their publicly released source code reproducing the results. See the last entry of the table below for an example.

Abbreviations

• FT = (classic) finetuning
• PT = prompt based tuning
• ICL = in-context learning, in the style of GPT-3
• μ±σ = mean μ and standard deviation σ across our 5 splits. Aggregate standard deviation is calculated using the sum-of-variance formula from individual tasks' standard deviations.

Benchmarking CLUES for Aggregate 30-shot Evaluation

Individual Task Performance over Multiple Shots

SST-2

MNLI

CoNLL03

WikiANN

SQuAD v2

ReCoRD

How do I cite CLUES?

@article{cluesteam2021,
  title={Few-Shot Learning Evaluation in Natural Language Understanding},
  author={Mukherjee, Subhabrata and Liu, Xiaodong and Zheng, Guoqing and Hosseini, Saghar and Cheng, Hao and Yang, Greg and Meek, Christopher and Awadallah, Ahmed Hassan and Gao, Jianfeng},
booktitle = {NeurIPS 2021},
year = {2021},
month = {December},
url = {https://www.microsoft.com/en-us/research/publication/clues-few-shot-learning-evaluation-in-natural-language-understanding/},
}

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.